MeridiaanShared context for every Agent
Meridiaan powers up the harness you already use. Every Agent you connect shares the same ideas, experiments and decisions over MCP, keeps them current and tells the others what it is doing: KnowledgeOps and AgentOps, in one place.
- 01
Organize your day.
- 1.Brainstorm the plan with your favorite Agent.
- 2.Save the Decision and the Todos.
- 3.Link each Todo to the spec it will need.
- 02
Split the work.
- 1.Your low-cost local Agent takes the simple Todos.
- 2.Your frontier Agent builds the demanding parts.
- 3.Both read the same Decision and tell each other what they touch.
- 03
Build on a teammate's work.
- 1.Your colleague's Agent writes the spec you need.
- 2.The spec is linked to your Todo.
- 3.Your Agents find it as soon as it is ready.
- 04
Pick up where you left off.
- 1.The next day, open a new session.
- 2.Your Agent finds the Decision, the open Todos and the finished spec.
- 3.Nothing to explain again.
What makes that possible
Six features, none of which you have to host.
Shared Knowledge
Documents live in Collections that declare what they hold and which statuses they allow. Each one is indexed as it changes, so an Agent pulls only what its task needs.
Learn more →Signals
Each Agent tells the others what it is working on and learns when a Document it read was rewritten, so two Agents do not step on each other.
Learn more →
Agent Sessions
Every call an Agent makes, session by session, with what it read and what it changed. You see all of it on the Agents page.
Learn more →Versioned Workbench
Commands, Skills and Loops written once in the Workspace, versioned, reach every Agent you connect without copying instructions into each repository.
Learn more →Works Everywhere
Each Workspace is an MCP server. Add its address to any of 50+ MCP clients and authorize it once in the browser with OAuth: nothing to deploy.
Learn more →Persona
Your own working preferences, written once and connected next to every Workspace, so every Agent works the way you like.
Learn more →Why not something you already have?
Why not a folder of Markdown files on my machine?
If you work alone with one Agent on one computer, files work, and they cost nothing. The picture changes when the work is spread across Agents: a frontier model for the hard problems, a local one for the easy ones, maybe a third in another terminal. Files on one machine reach only the Agents on that machine, and only the ones you pointed at that folder.
The second difference is the index. Whoever builds one over a folder has to keep it aligned, and a generic realignment means re-reading the whole knowledge base, so every write costs more as the project grows.
Why not an instructions file in the repository?
An instructions file is a good start, and it is free. It has two limits. It grows, and past a certain size it makes the Agent worse instead of better, because everything in it is loaded every time. And not everything a project knows belongs in a repository: decisions, discarded options, product and commercial context exist even when no code changes.
Developers also rarely use a single tool. A knowledge base that lives outside all of them is one that every tool can reach.
Context windows keep growing. Will I still need this?
Larger windows postpone the problem, they do not remove it. Chroma's Context Rot study (July 2025) tested eighteen frontier models and found that every one of them loses accuracy as the input grows, with serious degradation well before the declared window is full. Giving an Agent the few Documents it needs keeps it working on what matters.
Build with your Agents, from the same context.
Free plan available, no credit card required. Setup takes about five minutes.